AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

Security Analysis of Safe and Seldonian Reinforcement Learning Algorithms

Pinar Ozisik, Philip S. Thomas

NeurIPS 2020 · Reinforcement learning › Policy gradient and actor-critic · approach: Safety layers and shielding · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Panacea clipping to IS safety test, alpha=0.5CH, IS safety test · Diabetesadversarial trajectories needed: 49 -> 59Improved
add Panacea clipping to IS safety test, alpha=0.5CH, IS safety test · Grid-worldadversarial trajectories needed: 49 -> 68Improved
add Panacea clipping to WIS safety test, alpha=0.01CH, WIS safety test · Diabetesadversarial trajectories needed: 1 -> 65Improved
Panacea with IS at alpha=0.1CH, IS safety test · Both domainspasses unsafe policyImproved
Panacea with WIS on grid-worldCH, WIS safety test · Grid-worldpasses unsafe policyImproved
increase alpha from 0.5 to 5 in Panacea (IS)Panacea · Diabetesadversarial trajectories needed: 59 -> 49Hurt
increase alpha from 0.5 to 5 in Panacea (IS)Panacea · Grid-worldadversarial trajectories needed: 68 -> 50Hurt
increase alpha from 0.5 to 1 in Panacea (WIS)Panacea · Diabetesadversarial trajectories needed: 65 -> 1Hurt
alpha larger than alpha' of existing methodsPanacea · Both domainsadversarial trajectories neededNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy